Record Shadowing on Separate Tracks for a Clearer Comparison
Record model and shadowing voice on separate tracks, check latency, then compare timing without confusing recording delay with speaking delay.
Put the model on one track and your microphone on a second track, then check for a constant recording offset before judging your shadowing timing.
Before deciding that you are late, make sure the recording system is not late.
Why separate tracks make shadowing easier to compare
If the model and your voice are baked into one recording, comparison starts with a small detective story. Which voice was late? Was that missing syllable yours or buried under the source? Did the recorder introduce a delay?
Separate tracks remove much of that ambiguity. You can listen to the model alone, your voice alone, or both together. You can rerecord yourself without changing the source. And, crucially, you can check whether a timing difference comes from the recording setup before you treat it as a shadowing problem.
Audacity is one free example of a multitrack recorder. Its official overdubbing guide describes playing existing tracks while recording a new track and using track controls such as Mute and Solo. Other multitrack recorders can serve the same basic job. See Audacity's multitrack overdubbing guide.
Source: keep the model on its own track
Start with audio you are allowed to use: for example, your own recording, a licensed/downloadable learning file, or another lawful local copy. This page does not require—and should not become—a workaround for capturing protected streaming audio or bypassing DRM.
Import or place that audio on one track and label it clearly, such as MODEL. Leave the source track unchanged while you practise so you always have a stable reference.
Original practice example
I didn't realise it was going to take this long.
Was going to take presents a later event from a past viewpoint. “I didn't realise it will take this long” is understandable in some contexts, but it does not match that past viewpoint as naturally. Keep the model version on the source track as your unchanged reference.
Use this when something takes longer than you expected.
Label the source track MODEL and play the line once without recording. Make sure the wording and meaning are already clear before adding your track.
Explore more language-learning guides in Media-Based Language Learning.
Self: record your microphone onto a new track
Create a second track for your microphone and label it ME. In an overdub-style setup, the recorder plays the existing source while capturing your microphone onto the new track. The important part is not the software brand; it is keeping the two signals independent.
Use headphones when needed to prevent the source from leaking heavily back into the microphone. You do not need studio perfection. You need a learner track that is clear enough to review separately from the model.
Original practice example
Could you send it by Friday?
By Friday gives a completion deadline. Until Friday is grammatically valid for duration, but it does not express the same one-time deadline. Record your own version on the ME track while the model remains separate.
Use this as a polite deadline request at work or school.
Record one short attempt. Then solo MODEL and ME separately to confirm that each track can be heard on its own.
Sync: fix the ruler before measuring the learner
Digital recording systems can introduce latency: a recorded track may land slightly later than the sound you heard while performing. Audacity's current audio settings include latency compensation for this reason, and the amount can depend on your hardware and configuration. There is no honest universal number to subtract. See Audacity's current audio settings documentation.
This matters because a constant technical offset can look like a shadowing delay.
Original practice example
I didn't mean to interrupt.
Mean to + verb expresses intention. “I didn't mean interrupt” is wrong in standard English because to is required. If every phrase in your ME track appears shifted later by roughly the same amount—including this whole line—check recording latency before deciding that your shadowing timing is uniformly late.
Use this to apologise for an unintended interruption.
Compare the beginning and end of the recording. A similar whole-track offset points toward capture alignment; a delay that grows or appears only in certain phrases is different evidence.
Is the late timing technical or linguistic?
The whole learner track is shifted by about the same amount from start to finish.
Check latency or alignment first. A uniform offset is compatible with a recording-system delay. Correct or account for that before judging your performance.
The opening aligns, one phrase gets late, and I later recover.
That looks more like performance timing than one constant technical offset. Review the phrase where the timing changed rather than shifting the entire track.
The offset changes after I change my device, audio settings or recording setup.
Recheck the technical setup. Latency can depend on hardware and configuration, so an old compensation value may no longer fit.
The source and my voice are permanently mixed into one track.
Your comparison options are limited. You can still listen, but next time record the model and microphone separately so you can isolate each one.
The waveforms look different, but the speech sounds aligned.
Do not diagnose pronunciation from waveform shape alone. Use the audio and the language task, not visual similarity, as the main evidence.
Compare: model alone, yourself alone, then both
Once the tracks are technically trustworthy, comparison becomes much simpler:
- Solo the MODEL track and listen to one short phrase.
- Solo the ME track and listen to the same phrase.
- Play both together to hear the timing relationship.
- Change only playback levels if you need one track easier to hear; keep the original recordings intact.
British Council pronunciation guidance recommends recording yourself and comparing features such as sounds, rhythm, pace, stress and intonation. Separate tracks make that comparison easier to control, but they do not decide what matters most for you. See the British Council pronunciation guidance.
Original practice example
You don't have to decide right now.
Don't have to means there is no necessity. Mustn't is also grammatical, but it usually means prohibition. After the sync check, solo the model, solo your recording, then hear both together. Now a late right now can be reviewed as speech timing rather than an unexplained technical offset.
Use this to reassure someone that an immediate decision is unnecessary.
Compare just one phrase. If you find a real language target, stop the engineering work and practise the language.
Two tracks, not two dozen settings
You do not need to become an audio engineer to make shadowing comparison clearer. Keep the model separate. Keep yourself separate. Check sync. Then compare.
The important boundary is simple: fix the ruler before measuring the learner. Once a constant recording offset is ruled out, stop tweaking the recorder and return to the language.